The Agent War

I love Cold War TV shows, especially the ones involving spies. I’ve always been fascinated by the Soviet Union and that strange period of history when the U.S. and the USSR spent decades trying to outmaneuver each other without, you know, actually destroying the planet.

I’ve read books about it, watched more spy shows than is probably healthy (especially the ones featuring Keri Russell, who rocketed straight to the absolute top of my hall pass list during The Americans), and enjoyed games like Red Alert.

When I started reading about superpowers like the U.S. and China competing for RAM, chips, data centers, and computing power, my brain immediately started drawing parallels. This is what happens when you spend too much time watching Cold War trash and sometimes with inaccurate historical context.

If the Cold War was a competition over nuclear power, and the Space Race was a competition over who would reach the stars first, the Agent War is about who gets to build the machines that do the thinking, the working, and eventually, probably, the complaining for us.

And the thing is, like in the Space Race, it’s not really about who actually conquers the Moon. It’s about this fascination with throwing as many resources as possible into winning an imaginary race where nobody seems to be completely sure what the finish line is.

In reality, the battlefield looks considerably less exciting.

Even in the Cold War, there weren’t actually secret agents meeting in a dark parking garage to exchange microfilm every day. But still, instead of smokey Berlin alleyways we get executives discussing GPU supply chains in sterile, glass-walled silicon valley boardrooms.

Instead of launching rockets, we are building increasingly ridiculous data centers.

And instead of worrying about who has more nuclear warheads, we are apparently counting who builds the best model. Stealing models like trade secrets. Training bigger models, because apparently the answer to everything is "more GPUs."

We swapped nuclear proliferation for compute proliferation.

And the the more I look at it, the more the similarities start to pile up.

Several superpowers, especially the U.S. and China, are trying to lead the AI race. They need enormous amounts of resources to do it: chips, electricity, land, water, data centers, and, of course, people who know what the hell they are doing.

Instead of ICBM silos tucked into the Dakota plains, we are burying water-cooled mega-clusters in Oregon farmland, draining local rivers just so an LLM can generate synthetic memes.

And people are starting to push back against the infrastructure required to build all of this. We have protests around data centers today that remind me, at least superficially, of the public anxiety and protests surrounding nuclear power and nuclear weapons during the Cold War. We have our modern "hippies," with protests that range from the most ridiculous to the most seriously reflective—complete with homemade signs.

It’s Mutual Assured Destruction, but with a monthly subscription fee.

Then there is the really uncomfortable part: the fear of what happens if we lose control.

During the Cold War, scientists and governments worried about nuclear weapons getting out of control and eventually destroying civilization. We literally created the Doomsday Clock to communicate how close humanity might be to catastrophe.

Now we have scientists, researchers, and AI companies talking about increasingly capable systems becoming difficult to control, going rogue, or causing catastrophic damage.

Maybe we need a new Doomsday Clock.

Although I suppose instead of moving the hands toward midnight, we would have to move a little Microsoft Clippy icon toward some ominous singularity icon while Sam Altman and Jensen Huang insist that “if we don’t build the god-machine first, Beijing will.”

Just like the nuclear arms race, there is an interesting contradiction here: the same technology that everyone is afraid of is also the technology everyone wants to have.

Nobody wants to be the country without nuclear weapons if everyone else has them.

And nobody wants to be the company that says, "Maybe we should slow down," if everyone else keeps building more capable models—or more data centers.

That creates a strange feedback loop: fear becomes motivation, competition becomes justification, and suddenly everyone is running faster because everyone else is running faster.

Which brings me back to the Cold War.

Eventually, the U.S. and the Soviet Union had to sit down and talk. The 1985 Geneva Summit between Reagan and Gorbachev didn’t magically solve everything, but it helped establish a framework for continued negotiations. A couple of years later, they signed the INF Treaty.

Where is our Geneva Summit?

Not necessarily one giant summit where OpenAI, Anthropic, Microsoft, Google, Meta, SpaceX, and a bunch of exhausted government officials sit around a table and decide the future of humanity over coffee (and in the case of Elon Musk, ketamine).

But some kind of agreement about what we are willing to build, how we test it, how we deploy it, what we disclose, and where we draw some lines.

Because if this really is an Agent War, eventually someone has to figure out what the war is actually supposed to accomplish.

We are spending extraordinary amounts of money and resources to build machines that can increasingly think, reason, create, code, and act on our behalf. We are bankrupting the grid and melting silicon to build synthetic minds, racing at Mach 5 toward a finish line no one mapped out.

Duck under your desk. Put your head between your knees. Open the little agent in Microsoft Teams.

Welcome to the Agent War.